data

InceptionConfig

extendsModelConfig

InceptionConfig(num_classes: int = 1000, in_channels: int = 3, version: str = 'v3', aux_logits: bool = False, dropout: float = 0.5, transform_input: bool = False)

source edit

Configuration for Inception v3.

aux_logits enables the auxiliary classifier that attaches after the second InceptionC block (used during training with weight 0.4). dropout — main head dropout rate. 0.5 is the reference implementation's default, not the paper's: §8 (Training Methodology) specifies no dropout rate, and "dropout" appears in the text only in §4 and as "label-dropout" in §7. (timm defaults this to 0.0.) version — only "v3" is supported in this module.

Examples

>>> from lucid.models.vision.inception._config import InceptionConfig
>>> cfg = InceptionConfig()
>>> cfg.model_type
'inception_v3'
>>> cfg.num_classes, cfg.in_channels
(1000, 3)

Used by 3

Constructors

dunder

init

→None

__init__(num_classes: int = 1000, in_channels: int = 3, version: str = 'v3', aux_logits: bool = False, dropout: float = 0.5, transform_input: bool = False)

source edit

data

InceptionConfig

extendsModelConfig

InceptionConfig(num_classes: int = 1000, in_channels: int = 3, version: str = 'v3', aux_logits: bool = False, dropout: float = 0.5, transform_input: bool = False)

source edit

Configuration for Inception v3.

Examples

>>> from lucid.models.vision.inception._config import InceptionConfig
>>> cfg = InceptionConfig()
>>> cfg.model_type
'inception_v3'
>>> cfg.num_classes, cfg.in_channels
(1000, 3)

Used by 3

Constructors

dunder

init

→None

__init__(num_classes: int = 1000, in_channels: int = 3, version: str = 'v3', aux_logits: bool = False, dropout: float = 0.5, transform_input: bool = False)

source edit